integrity of research data data acquisition, management, and sharing

Integrity of Research Data

Data Acquisition, Management, and Sharing

Definition of DATA

Recorded information; includes writings, films, sound recordings, pictorial reproductions,

drawings, designs, or othergraphic representations, procedural manuals, forms, diagrams, work flow charts, equipment

descriptions, data files, data processing or computer programs (software), statistical

records, and other research data[NIH Grants Policy Statement,

“Rights in Data”]

Some Issues In Data Integrity

Acquisition and record keeping Processing Ownership and control Retention and storage Access and sharing

Policies & GuidelinesConcerning Data Integrity Federal (e.g., Public Health Service, Food

and Drug Administration) Institutional (e.g., MIT) Laboratories or research groups

Data Acquisition andRecord-Keeping

Record Keeping

Standards from FDA (standards for eventual approval of biologics, drugs, devices)

• Basic science: “Good Laboratory Practices”or GLP’s• Clinical studies: “Good Clinical Practices” orGCP’s• Standards for design, conduct, recording, reporting of research studies – data for approval of use must be of “highest quality and integrity” [FDA]

Record Keeping in Research Studies FDA GLP:• All data should be recorded promptly• Entries must be dated and signed• Any changes should not obscure original entry• Reason for change must be indicated, signed,

and dated

[21 CFR PART 58]

Laboratory Notebooks

What to record?• Methods• Results/observations• Ideas• Location of relevant computer & paper files• Changes to protocol• Notes from lab meetings• Contributors at each stage

Laboratory Notebooks

How to record?

• As results are generated

• Chronologically

• With clear, readable entries

Laboratory NotebooksDo’s and Don’ts

Do

• Use bound notebook with sequentially numbered pages

• Write in black ink (resistant to moisture, solvents)

• Sign and date each entry

• Correct mistakes by slash, initialed and reason given


Do

• Include attachments by taping into notebook and putting initials/date across tape

• Save notebooks


Don’t

• Erase

• Leave blank spaces without slash, signature, and date

• Tear out pages

Record Keeping in Clinical Studies

FDA GCP• Maintain case history records

Case report forms Supporting documents

• Ensure accuracy, completeness, legibility, and timeliness of data

• Make corrections to case history records so as not to obscure original entry (applies to both written and electronic corrections)

[Federal Register, vol. 60, number 159]

Record Keeping inClinical Studies Case History Records should include

• Basic subject information• Information that subject meets eligibility criteria• Information on treatment• Information from tests and examinations

Lab results X-rays Physical examinations, etc.

Form of record can be printed, optical, or electronic

Record KeepingUsing ComputersComputerized data collection systems must

• Allow data entry by authorized individuals only• Control ability to delete or alter entries• Provide audit trail

Who made change When Why

• Protect from tampering• Ensure preservation of data (backup & recovery)• Have system for generating print-outs

Data Processing

Data Processing

Maximize following in data set:• Completeness• Accuracy

Verify that• Data are entered correctly• Each variable falls within proper range• No missing values (or identify and handle

cases with missing information)

Data Entry

Use double entry systems• Enter observations into initial data set• Re-enter to validate

Use direct transfer from equipment Check after entry

Out-of-Range Values

“Outlier” is observation that appears inconsistent with remainder of data set

Detect outliers in a single variable• Are the minimum and maximum values for

each variable what you expect?• Does the distribution have outlying tails?

If possible, determine cause of outliers Decide how to deal with them before

analysis

Possible Sources ofOutliers Errors made in taking, recording, or

entering data Case with extreme value is not part of

population you intended to sample Extreme value is result of extreme (but

real) biological or environmental variation

Example

Case F (N)1 202 34003 45004 2010… …16 7000

Example

Example

When checking original tracing of load

cell output, find force for Case 1 = 2000 N

data entry error When checking original description of

cases, find that Case 16 is male. Sample was intended to be female Æ case is not part of intended sample

Outliers in a Combination ofVariables Unusual combination of two or more

measurements Can use scatter plots to search among

continuous variables for unusual patterns There are statistical multivariate methods

to detect these outliers

Example of Multivariate Outlier

Outlying DataWhat To Do

Strategy depends on source Inaccurate data entry

• Replace with correct value

Case is not from target population• Delete case if you know that you inadvertently

sampled unit that is not from target population• Run analysis with and without case and

compare results if you only suspect

Outlying DataWhat To Do Extreme variation

• Run analysis with and without case and compare results

• Use techniques during the analysis phase that adjust for skewed data

• Live with it and accept any distortion caused by outlier

Missing Values

Identify and flag missing values Determine cause if possible

• Examples: Patient drop-out, machine malfunction

Quantify amount of missing information Check pattern -- should be random Decide how to deal with missing values

before analysis

Missing Values -- What To Do

Drop variable with missing value(s) Drop case with missing value(s) Accept unequal numbers of observation Be cautious with repeated-measures design Bone Mineral Density (g/cm2)


Impute missing values (be careful)• Use average of non-missing values

Potential problem: variability of new data set may be underestimated

• Use preceding non-missing value after sorting data into meaningful order

• Use relationships with other variables to estimate missing value (do not use variables to be used in hypothesis testing)


Treat missing data as data• Use when failure to have value may itself be

predictor of outcome in your study• Make new “dummy” variable out of variable

with missing values and use as another variable in analysis

Complete = 0 Missing = 1

Data Processing -- Summary

Data Processing – Responsible Conduct

Refining data set may require omission or modification

Disclose basis for dropping or modifying data

Data Ownership and Custody

Data Ownership

Data ownership standards and policies depend on source of support• Public Health Service (PHS): Data generated

under PHS grant is owned by grantee institution

• Sponsored research: Ownership and control depend on research contract

• Unsupported: Typically institution retains ownership

Data Ownership – PHS Policy

“In general, grantees own the data generated by or resulting from a grant- supported project”

NIH Grants Policy Statement,

“Rights in Data”

Data Control - SponsoredResearch For sponsored research contracts, control

of data is often in “publication” paragraph Examples:

• Institution is free to publish results of research after providing Sponsor with time to review for Patentable material Inadvertent disclosure of proprietary information

Data Custody

Custody or guardianship is typically in hands of PI or Laboratory (on behalf of institution)

Departure of Investigator

Policies regarding departure depend on institution

Examples:• If PI leaves institution

PI able to remove original data while allowing future access

PI takes copies

• If co-investigator, post-doc, student leaves Original data left with PI; copies provided Original data left with PI; access allowed

Data Retention and Storage

How long must original data records be retained?

• PHS: 3 years after end of grant

[45 CFR Part 74]

• FDA: 2 years after marketing approval

[21 CFR Part 58]

• Institutions/laboratories: Varies from 2 years after publication to indefinitely


Where should original data records be stored?

• Often not stated explicitly

• Examples: In laboratory In research office of institution


Under what conditions should original data be stored?

• Ideal conditions Temperature and humidity controlled Protection against natural disaster or theft Access controlled (no alterations) but easy to

retrieve

• Reality File drawer

Documentation of Storage

Usually no set institutional policy Recommendation: Maintain tracking directory

that records• Type of data (notebook, tape, x-ray, etc.)• and identifying number• Project title• Investigator• Date data were recorded or notebook completed• Storage location

Data Sharing

Data Access and Sharing

Potentially opposing goals of data exchange:

• Right of investigators to get credit for data

(through analysis, presentation at research meetings, publication)

versus

• Obligation of investigators to make data available to colleagues

Data Access and Sharing

Why do we need access?

• Openness and exchange of information are basic tenets of science

• Access may be required to investigate allegations of misconduct

• Access may be requested through Freedom of Information Act

• Access required for PHS-funded research

MIT Policy Statement on DataSharing

“The prompt and open dissemination of the results of M.I.T. research and the free exchange of information among scholars are essential to the fulfillment of M.I.T.'s obligations as an institution committed to excellence in education and research”

Guide to the Ownership, Distribution and

Commercial Development of M.I.T. Technology

NIH Statement on SharingResearch Data NIH expects timely release and sharing of

final research data for use by other researchers

NIH will require applicants to include a plan for data sharing or to state why data sharing is not possible. (applies to grants over certain amount)

NIH Statement: Why share?

Extends NIH policy on sharing research resources

Reinforces open scientific inquiry Encourages diversity of analysis and opinion Promotes new research Supports testing of new or alternative

hypotheses and methods of analysis

Why share?

Facilitates the education of new researchers

Enables the exploration of topics not envisioned by the initial investigators

Permits the creation of new data sets from combined data

What kind of information should be shared?

Final research data necessary to validate research findings

Does not include:• laboratory notebooks• partial data sets• preliminary analyses• drafts of scientific papers• plans for future research• communications with colleagues• physical objects, such as gels or lab specimens

To what kind of research does this apply?

Data generated with support from NIH:• Basic research• Clinical studies• Surveys• Other types of research

Especially important to share:• Unique data sets that cannot be readily

replicated• Large, expensive data sets

Caveats for studies including human research participants Investigators need to be cautious with

• Studies with very small samples• Studies collecting very sensitive data

However, even these data can be shared if safeguards exist to ensure confidentiality and protect identify of subjects

What is meant by timely?

No later than acceptance for publication of main findings from the final data set

How to share data?

Provide in publications Share under the investigator’s own

auspices Place data sets in public archives Put data on a web site Place in restricted access data centers or

data enclaves Other ways

Summary

Summary

Policies on responsible data management are evolving

Some current views:• Investigator’s responsibility to ensure

accurate and reliable data set, including Primary sources (e.g., notebooks) Refined data set

Summary

• Institutions tend to own data• PI’s or laboratories tend to have custody of data• Original data should be retained for ≥3 years• Access is required for PHS-funded research

(after publication) and may be required for other types of support

• Open dissemination of results is basic tenet of research community

Resources

Committee on National Statistics, National Research Council, Sharing Research Data, National Academies Press, available at http://www.nap.edu/books/030903499X/html/index.html

Resources

MIT TLO Policy Guide, available at http://web.mit.edu/tlo/www/guide.2.html

National Aeronautics and Space Administration(NASA) Guidelines for Ensuring the Quality of

Information, available at ftp://ftp.hq.nasa.gov/pub/pao/reports/2002/NAS

A_data_quality_guidelines.pdf NIH Data Sharing Information Main Page, available at

http://grants2.nih.gov/grants/policy/data_sharing/

integrity of research data data acquisition, management, and sharing

Documents